xref: /freebsd/sys/kern/kern_shutdown.c (revision e453e498cbb88570a3ff7b3679de65c88707da95)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1986, 1988, 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include <sys/cdefs.h>
38 #include "opt_ddb.h"
39 #include "opt_ekcd.h"
40 #include "opt_kdb.h"
41 #include "opt_panic.h"
42 #include "opt_printf.h"
43 #include "opt_sched.h"
44 #include "opt_watchdog.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/bio.h>
49 #include <sys/boottrace.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/compressor.h>
53 #include <sys/cons.h>
54 #include <sys/disk.h>
55 #include <sys/eventhandler.h>
56 #include <sys/filedesc.h>
57 #include <sys/jail.h>
58 #include <sys/kdb.h>
59 #include <sys/kernel.h>
60 #include <sys/kerneldump.h>
61 #include <sys/kthread.h>
62 #include <sys/ktr.h>
63 #include <sys/malloc.h>
64 #include <sys/mbuf.h>
65 #include <sys/mount.h>
66 #include <sys/priv.h>
67 #include <sys/proc.h>
68 #include <sys/reboot.h>
69 #include <sys/resourcevar.h>
70 #include <sys/rwlock.h>
71 #include <sys/sbuf.h>
72 #include <sys/sched.h>
73 #include <sys/smp.h>
74 #include <sys/stdarg.h>
75 #include <sys/sysctl.h>
76 #include <sys/sysproto.h>
77 #include <sys/taskqueue.h>
78 #include <sys/vnode.h>
79 #include <sys/watchdog.h>
80 
81 #include <crypto/chacha20/chacha.h>
82 #include <crypto/rijndael/rijndael-api-fst.h>
83 #include <crypto/sha2/sha256.h>
84 
85 #include <ddb/ddb.h>
86 
87 #include <machine/cpu.h>
88 #include <machine/dump.h>
89 #include <machine/pcb.h>
90 #include <machine/smp.h>
91 
92 #include <security/mac/mac_framework.h>
93 
94 #include <vm/vm.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_page.h>
97 #include <vm/vm_pager.h>
98 #include <vm/swap_pager.h>
99 
100 #include <sys/signalvar.h>
101 
102 static MALLOC_DEFINE(M_DUMPER, "dumper", "dumper block buffer");
103 
104 #ifndef PANIC_REBOOT_WAIT_TIME
105 #define PANIC_REBOOT_WAIT_TIME 15 /* default to 15 seconds */
106 #endif
107 static int panic_reboot_wait_time = PANIC_REBOOT_WAIT_TIME;
108 SYSCTL_INT(_kern, OID_AUTO, panic_reboot_wait_time, CTLFLAG_RWTUN,
109     &panic_reboot_wait_time, 0,
110     "Seconds to wait before rebooting after a panic");
111 static int reboot_wait_time = 0;
112 SYSCTL_INT(_kern, OID_AUTO, reboot_wait_time, CTLFLAG_RWTUN,
113     &reboot_wait_time, 0,
114     "Seconds to wait before rebooting");
115 
116 #ifdef KDB
117 #ifdef KDB_UNATTENDED
118 int debugger_on_panic = 0;
119 #else
120 int debugger_on_panic = 1;
121 #endif
122 SYSCTL_INT(_debug, OID_AUTO, debugger_on_panic,
123     CTLFLAG_RWTUN, &debugger_on_panic, 0,
124     "Run debugger on kernel panic");
125 
126 static bool debugger_on_recursive_panic = false;
127 SYSCTL_BOOL(_debug, OID_AUTO, debugger_on_recursive_panic,
128     CTLFLAG_RWTUN, &debugger_on_recursive_panic, 0,
129     "Run debugger on recursive kernel panic");
130 
131 int debugger_on_trap = 0;
132 SYSCTL_INT(_debug, OID_AUTO, debugger_on_trap,
133     CTLFLAG_RWTUN, &debugger_on_trap, 0,
134     "Run debugger on kernel trap before panic");
135 
136 #ifdef KDB_TRACE
137 static int trace_on_panic = 1;
138 static bool trace_all_panics = true;
139 #else
140 static int trace_on_panic = 0;
141 static bool trace_all_panics = false;
142 #endif
143 SYSCTL_INT(_debug, OID_AUTO, trace_on_panic,
144     CTLFLAG_RWTUN | CTLFLAG_SECURE,
145     &trace_on_panic, 0, "Print stack trace on kernel panic");
146 SYSCTL_BOOL(_debug, OID_AUTO, trace_all_panics, CTLFLAG_RWTUN,
147     &trace_all_panics, 0, "Print stack traces on secondary kernel panics");
148 #endif /* KDB */
149 
150 static int sync_on_panic = 0;
151 SYSCTL_INT(_kern, OID_AUTO, sync_on_panic, CTLFLAG_RWTUN,
152 	&sync_on_panic, 0, "Do a sync before rebooting from a panic");
153 
154 static bool poweroff_on_panic = 0;
155 SYSCTL_BOOL(_kern, OID_AUTO, poweroff_on_panic, CTLFLAG_RWTUN,
156 	&poweroff_on_panic, 0, "Do a power off instead of a reboot on a panic");
157 
158 static bool powercycle_on_panic = 0;
159 SYSCTL_BOOL(_kern, OID_AUTO, powercycle_on_panic, CTLFLAG_RWTUN,
160 	&powercycle_on_panic, 0, "Do a power cycle instead of a reboot on a panic");
161 
162 static SYSCTL_NODE(_kern, OID_AUTO, shutdown, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
163     "Shutdown environment");
164 
165 #ifndef DIAGNOSTIC
166 static int show_busybufs;
167 #else
168 static int show_busybufs = 1;
169 #endif
170 SYSCTL_INT(_kern_shutdown, OID_AUTO, show_busybufs, CTLFLAG_RW,
171     &show_busybufs, 0,
172     "Show busy buffers during shutdown");
173 
174 int suspend_blocked = 0;
175 SYSCTL_INT(_kern, OID_AUTO, suspend_blocked, CTLFLAG_RW,
176 	&suspend_blocked, 0, "Block suspend due to a pending shutdown");
177 
178 #ifdef EKCD
179 FEATURE(ekcd, "Encrypted kernel crash dumps support");
180 
181 MALLOC_DEFINE(M_EKCD, "ekcd", "Encrypted kernel crash dumps data");
182 
183 struct kerneldumpcrypto {
184 	uint8_t			kdc_encryption;
185 	uint8_t			kdc_iv[KERNELDUMP_IV_MAX_SIZE];
186 	union {
187 		struct {
188 			keyInstance	aes_ki;
189 			cipherInstance	aes_ci;
190 		} u_aes;
191 		struct chacha_ctx	u_chacha;
192 	} u;
193 #define	kdc_ki	u.u_aes.aes_ki
194 #define	kdc_ci	u.u_aes.aes_ci
195 #define	kdc_chacha	u.u_chacha
196 	uint32_t		kdc_dumpkeysize;
197 	struct kerneldumpkey	kdc_dumpkey[];
198 };
199 #endif
200 
201 struct kerneldumpcomp {
202 	uint8_t			kdc_format;
203 	struct compressor	*kdc_stream;
204 	uint8_t			*kdc_buf;
205 	size_t			kdc_resid;
206 };
207 
208 static struct kerneldumpcomp *kerneldumpcomp_create(struct dumperinfo *di,
209 		    uint8_t compression);
210 static void	kerneldumpcomp_destroy(struct dumperinfo *di);
211 static int	kerneldumpcomp_write_cb(void *base, size_t len, off_t off, void *arg);
212 
213 static int kerneldump_gzlevel = 6;
214 SYSCTL_INT(_kern, OID_AUTO, kerneldump_gzlevel, CTLFLAG_RWTUN,
215     &kerneldump_gzlevel, 0,
216     "Kernel crash dump compression level");
217 
218 /*
219  * Variable panicstr contains argument to first call to panic; used as flag
220  * to indicate that the kernel has already called panic.
221  */
222 const char *panicstr __read_mostly;
223 bool scheduler_stopped __read_frequently;
224 
225 int dumping __read_mostly;		/* system is dumping */
226 int rebooting __read_mostly;		/* system is rebooting */
227 bool dumped_core __read_mostly;		/* system successfully dumped core */
228 /*
229  * Used to serialize between sysctl kern.shutdown.dumpdevname and list
230  * modifications via ioctl.
231  */
232 static struct mtx dumpconf_list_lk;
233 MTX_SYSINIT(dumper_configs, &dumpconf_list_lk, "dumper config list", MTX_DEF);
234 
235 /* Our selected dumper(s). */
236 static TAILQ_HEAD(dumpconflist, dumperinfo) dumper_configs =
237     TAILQ_HEAD_INITIALIZER(dumper_configs);
238 
239 /* Context information for dump-debuggers, saved by the dump_savectx() macro. */
240 struct pcb dumppcb;			/* Registers. */
241 lwpid_t dumptid;			/* Thread ID. */
242 
243 static struct cdevsw reroot_cdevsw = {
244      .d_version = D_VERSION,
245      .d_name    = "reroot",
246 };
247 
248 static void poweroff_wait(void *, int);
249 static void shutdown_halt(void *junk, int howto);
250 static void shutdown_panic(void *junk, int howto);
251 static void shutdown_reset(void *junk, int howto);
252 static int kern_reroot(void);
253 
254 /* register various local shutdown events */
255 static void
shutdown_conf(void * unused)256 shutdown_conf(void *unused)
257 {
258 
259 	EVENTHANDLER_REGISTER(shutdown_final, poweroff_wait, NULL,
260 	    SHUTDOWN_PRI_FIRST);
261 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_panic, NULL,
262 	    SHUTDOWN_PRI_LAST + 100);
263 	EVENTHANDLER_REGISTER(shutdown_final, shutdown_halt, NULL,
264 	    SHUTDOWN_PRI_LAST + 200);
265 }
266 
267 SYSINIT(shutdown_conf, SI_SUB_INTRINSIC, SI_ORDER_ANY, shutdown_conf, NULL);
268 
269 /*
270  * The only reason this exists is to create the /dev/reroot/ directory,
271  * used by reroot code in init(8) as a mountpoint for tmpfs.
272  */
273 static void
reroot_conf(void * unused)274 reroot_conf(void *unused)
275 {
276 	int error;
277 	struct cdev *cdev;
278 
279 	error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &cdev,
280 	    &reroot_cdevsw, NULL, UID_ROOT, GID_WHEEL, 0600, "reroot/reroot");
281 	if (error != 0) {
282 		printf("%s: failed to create device node, error %d",
283 		    __func__, error);
284 	}
285 }
286 
287 SYSINIT(reroot_conf, SI_SUB_DEVFS, SI_ORDER_ANY, reroot_conf, NULL);
288 
289 /*
290  * The system call that results in a reboot.
291  */
292 /* ARGSUSED */
293 int
sys_reboot(struct thread * td,struct reboot_args * uap)294 sys_reboot(struct thread *td, struct reboot_args *uap)
295 {
296 	int error;
297 
298 	error = 0;
299 #ifdef MAC
300 	error = mac_system_check_reboot(td->td_ucred, uap->opt);
301 #endif
302 	if (error == 0)
303 		error = priv_check(td, PRIV_REBOOT);
304 	if (error == 0) {
305 		if (uap->opt & RB_REROOT)
306 			error = kern_reroot();
307 		else
308 			kern_reboot(uap->opt);
309 	}
310 	return (error);
311 }
312 
313 static void
shutdown_nice_task_fn(void * arg,int pending __unused)314 shutdown_nice_task_fn(void *arg, int pending __unused)
315 {
316 	int howto;
317 
318 	howto = (uintptr_t)arg;
319 	/* Send a signal to init(8) and have it shutdown the world. */
320 	PROC_LOCK(initproc);
321 	if ((howto & RB_POWEROFF) != 0) {
322 		BOOTTRACE("SIGUSR2 to init(8)");
323 		kern_psignal(initproc, SIGUSR2);
324 	} else if ((howto & RB_POWERCYCLE) != 0) {
325 		BOOTTRACE("SIGWINCH to init(8)");
326 		kern_psignal(initproc, SIGWINCH);
327 	} else if ((howto & RB_HALT) != 0) {
328 		BOOTTRACE("SIGUSR1 to init(8)");
329 		kern_psignal(initproc, SIGUSR1);
330 	} else {
331 		BOOTTRACE("SIGINT to init(8)");
332 		kern_psignal(initproc, SIGINT);
333 	}
334 	PROC_UNLOCK(initproc);
335 }
336 
337 static struct task shutdown_nice_task = TASK_INITIALIZER(0,
338     &shutdown_nice_task_fn, NULL);
339 
340 /*
341  * Called by events that want to shut down.. e.g  <CTL><ALT><DEL> on a PC
342  */
343 void
shutdown_nice(int howto)344 shutdown_nice(int howto)
345 {
346 
347 	if (initproc != NULL && !SCHEDULER_STOPPED()) {
348 		BOOTTRACE("shutdown initiated");
349 		shutdown_nice_task.ta_context = (void *)(uintptr_t)howto;
350 		taskqueue_enqueue(taskqueue_fast, &shutdown_nice_task);
351 	} else {
352 		/*
353 		 * No init(8) running, or scheduler would not allow it
354 		 * to run, so simply reboot.
355 		 */
356 		kern_reboot(howto | RB_NOSYNC);
357 	}
358 }
359 
360 static void
print_uptime(void)361 print_uptime(void)
362 {
363 	int f;
364 	struct timespec ts;
365 
366 	getnanouptime(&ts);
367 	printf("Uptime: ");
368 	f = 0;
369 	if (ts.tv_sec >= 86400) {
370 		printf("%ldd", (long)ts.tv_sec / 86400);
371 		ts.tv_sec %= 86400;
372 		f = 1;
373 	}
374 	if (f || ts.tv_sec >= 3600) {
375 		printf("%ldh", (long)ts.tv_sec / 3600);
376 		ts.tv_sec %= 3600;
377 		f = 1;
378 	}
379 	if (f || ts.tv_sec >= 60) {
380 		printf("%ldm", (long)ts.tv_sec / 60);
381 		ts.tv_sec %= 60;
382 		f = 1;
383 	}
384 	printf("%lds\n", (long)ts.tv_sec);
385 }
386 
387 int
doadump(boolean_t textdump)388 doadump(boolean_t textdump)
389 {
390 	boolean_t coredump;
391 	int error;
392 
393 	error = 0;
394 	if (dumping)
395 		return (EBUSY);
396 	if (TAILQ_EMPTY(&dumper_configs))
397 		return (ENXIO);
398 
399 	dump_savectx();
400 	dumping++;
401 
402 	coredump = TRUE;
403 #ifdef DDB
404 	if (textdump && textdump_pending) {
405 		coredump = FALSE;
406 		textdump_dumpsys(TAILQ_FIRST(&dumper_configs));
407 	}
408 #endif
409 	if (coredump) {
410 		struct dumperinfo *di;
411 
412 		TAILQ_FOREACH(di, &dumper_configs, di_next) {
413 			error = dumpsys(di);
414 			if (error == 0) {
415 				dumped_core = true;
416 				break;
417 			}
418 		}
419 	}
420 
421 	dumping--;
422 	return (error);
423 }
424 
425 /*
426  * Trace the shutdown reason.
427  */
428 static void
reboottrace(int howto)429 reboottrace(int howto)
430 {
431 	if ((howto & RB_DUMP) != 0) {
432 		if ((howto & RB_HALT) != 0)
433 			BOOTTRACE("system panic: halting...");
434 		if ((howto & RB_POWEROFF) != 0)
435 			BOOTTRACE("system panic: powering off...");
436 		if ((howto & (RB_HALT|RB_POWEROFF)) == 0)
437 			BOOTTRACE("system panic: rebooting...");
438 	} else {
439 		if ((howto & RB_HALT) != 0)
440 			BOOTTRACE("system halting...");
441 		if ((howto & RB_POWEROFF) != 0)
442 			BOOTTRACE("system powering off...");
443 		if ((howto & (RB_HALT|RB_POWEROFF)) == 0)
444 			BOOTTRACE("system rebooting...");
445 	}
446 }
447 
448 /*
449  * kern_reboot(9): Shut down the system cleanly to prepare for reboot, halt, or
450  * power off.
451  */
452 void
kern_reboot(int howto)453 kern_reboot(int howto)
454 {
455 	static int once = 0;
456 
457 	if (initproc != NULL && curproc != initproc)
458 		BOOTTRACE("kernel shutdown (dirty) started");
459 	else
460 		BOOTTRACE("kernel shutdown (clean) started");
461 
462 	/*
463 	 * Normal paths here don't hold Giant, but we can wind up here
464 	 * unexpectedly with it held.  Drop it now so we don't have to
465 	 * drop and pick it up elsewhere. The paths it is locking will
466 	 * never be returned to, and it is preferable to preclude
467 	 * deadlock than to lock against code that won't ever
468 	 * continue.
469 	 */
470 	while (!SCHEDULER_STOPPED() && mtx_owned(&Giant))
471 		mtx_unlock(&Giant);
472 
473 #if defined(SMP)
474 	/*
475 	 * Bind us to the first CPU so that all shutdown code runs there.  Some
476 	 * systems don't shutdown properly (i.e., ACPI power off) if we
477 	 * run on another processor.
478 	 */
479 	if (!SCHEDULER_STOPPED()) {
480 		thread_lock(curthread);
481 		sched_bind(curthread, CPU_FIRST());
482 		thread_unlock(curthread);
483 		KASSERT(PCPU_GET(cpuid) == CPU_FIRST(),
484 		    ("%s: not running on cpu 0", __func__));
485 	}
486 #endif
487 	/* We're in the process of rebooting. */
488 	rebooting = 1;
489 	reboottrace(howto);
490 
491 	/*
492 	 * Do any callouts that should be done BEFORE syncing the filesystems.
493 	 */
494 	EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
495 	BOOTTRACE("shutdown pre sync complete");
496 
497 	/*
498 	 * Now sync filesystems
499 	 */
500 	if (!cold && (howto & RB_NOSYNC) == 0 && once == 0) {
501 		once = 1;
502 		BOOTTRACE("bufshutdown begin");
503 		bufshutdown(show_busybufs);
504 		BOOTTRACE("bufshutdown end");
505 	}
506 
507 	print_uptime();
508 
509 	cngrab();
510 
511 	/*
512 	 * Ok, now do things that assume all filesystem activity has
513 	 * been completed.
514 	 */
515 	EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
516 	BOOTTRACE("shutdown post sync complete");
517 
518 	if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
519 		doadump(TRUE);
520 
521 	/* Now that we're going to really halt the system... */
522 	BOOTTRACE("shutdown final begin");
523 
524 	if (shutdown_trace)
525 		boottrace_dump_console();
526 
527 	EVENTHANDLER_INVOKE(shutdown_final, howto);
528 
529 	/*
530 	 * Call this directly so that reset is attempted even if shutdown
531 	 * handlers are not yet registered.
532 	 */
533 	shutdown_reset(NULL, howto);
534 
535 	for(;;) ;	/* safety against shutdown_reset not working */
536 	/* NOTREACHED */
537 }
538 
539 /*
540  * The system call that results in changing the rootfs.
541  */
542 static int
kern_reroot(void)543 kern_reroot(void)
544 {
545 	struct vnode *oldrootvnode, *vp;
546 	struct mount *mp, *devmp;
547 	int error;
548 
549 	if (curproc != initproc)
550 		return (EPERM);
551 
552 	/*
553 	 * Mark the filesystem containing currently-running executable
554 	 * (the temporary copy of init(8)) busy.
555 	 */
556 	vp = curproc->p_textvp;
557 	error = vn_lock(vp, LK_SHARED);
558 	if (error != 0)
559 		return (error);
560 	mp = vp->v_mount;
561 	error = vfs_busy(mp, MBF_NOWAIT);
562 	if (error != 0) {
563 		vfs_ref(mp);
564 		VOP_UNLOCK(vp);
565 		error = vfs_busy(mp, 0);
566 		vn_lock(vp, LK_SHARED | LK_RETRY);
567 		vfs_rel(mp);
568 		if (error != 0) {
569 			VOP_UNLOCK(vp);
570 			return (ENOENT);
571 		}
572 		if (VN_IS_DOOMED(vp)) {
573 			VOP_UNLOCK(vp);
574 			vfs_unbusy(mp);
575 			return (ENOENT);
576 		}
577 	}
578 	VOP_UNLOCK(vp);
579 
580 	/*
581 	 * Remove the filesystem containing currently-running executable
582 	 * from the mount list, to prevent it from being unmounted
583 	 * by vfs_unmountall(), and to avoid confusing vfs_mountroot().
584 	 *
585 	 * Also preserve /dev - forcibly unmounting it could cause driver
586 	 * reinitialization.
587 	 */
588 
589 	vfs_ref(rootdevmp);
590 	devmp = rootdevmp;
591 	rootdevmp = NULL;
592 
593 	mtx_lock(&mountlist_mtx);
594 	TAILQ_REMOVE(&mountlist, mp, mnt_list);
595 	TAILQ_REMOVE(&mountlist, devmp, mnt_list);
596 	mtx_unlock(&mountlist_mtx);
597 
598 	oldrootvnode = rootvnode;
599 
600 	/*
601 	 * Unmount everything except for the two filesystems preserved above.
602 	 */
603 	vfs_unmountall();
604 
605 	/*
606 	 * Add /dev back; vfs_mountroot() will move it into its new place.
607 	 */
608 	mtx_lock(&mountlist_mtx);
609 	TAILQ_INSERT_HEAD(&mountlist, devmp, mnt_list);
610 	mtx_unlock(&mountlist_mtx);
611 	rootdevmp = devmp;
612 	vfs_rel(rootdevmp);
613 
614 	/*
615 	 * Mount the new rootfs.
616 	 */
617 	vfs_mountroot();
618 
619 	/*
620 	 * Update all references to the old rootvnode.
621 	 */
622 	mountcheckdirs(oldrootvnode, rootvnode);
623 
624 	/*
625 	 * Add the temporary filesystem back and unbusy it.
626 	 */
627 	mtx_lock(&mountlist_mtx);
628 	TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
629 	mtx_unlock(&mountlist_mtx);
630 	vfs_unbusy(mp);
631 
632 	return (0);
633 }
634 
635 /*
636  * If the shutdown was a clean halt, behave accordingly.
637  */
638 static void
shutdown_halt(void * junk,int howto)639 shutdown_halt(void *junk, int howto)
640 {
641 
642 	if (howto & RB_HALT) {
643 		printf("\n");
644 		printf("The operating system has halted.\n");
645 		printf("Please press any key to reboot.\n\n");
646 
647 		wdog_kern_pat(WD_TO_NEVER);
648 
649 		switch (cngetc()) {
650 		case -1:		/* No console, just die */
651 			cpu_halt();
652 			/* NOTREACHED */
653 		default:
654 			break;
655 		}
656 	}
657 }
658 
659 /*
660  * Check to see if the system panicked, pause and then reboot
661  * according to the specified delay.
662  */
663 static void
shutdown_panic(void * junk,int howto)664 shutdown_panic(void *junk, int howto)
665 {
666 	int loop;
667 
668 	if (howto & RB_DUMP) {
669 		if (panic_reboot_wait_time != 0) {
670 			if (panic_reboot_wait_time != -1) {
671 				printf("Automatic reboot in %d seconds - "
672 				       "press a key on the console to abort\n",
673 					panic_reboot_wait_time);
674 				for (loop = panic_reboot_wait_time * 10;
675 				     loop > 0; --loop) {
676 					DELAY(1000 * 100); /* 1/10th second */
677 					/* Did user type a key? */
678 					if (cncheckc() != -1)
679 						break;
680 				}
681 				if (!loop)
682 					return;
683 			}
684 		} else { /* zero time specified - reboot NOW */
685 			return;
686 		}
687 		printf("--> Press a key on the console to reboot,\n");
688 		printf("--> or switch off the system now.\n");
689 		cngetc();
690 	}
691 }
692 
693 /*
694  * Everything done, now reset
695  */
696 static void
shutdown_reset(void * junk,int howto)697 shutdown_reset(void *junk, int howto)
698 {
699 
700 	printf("Rebooting...\n");
701 	DELAY(reboot_wait_time * 1000000);
702 
703 	/*
704 	 * Acquiring smp_ipi_mtx here has a double effect:
705 	 * - it disables interrupts avoiding CPU0 preemption
706 	 *   by fast handlers (thus deadlocking  against other CPUs)
707 	 * - it avoids deadlocks against smp_rendezvous() or, more
708 	 *   generally, threads busy-waiting, with this spinlock held,
709 	 *   and waiting for responses by threads on other CPUs
710 	 *   (ie. smp_tlb_shootdown()).
711 	 *
712 	 * For the !SMP case it just needs to handle the former problem.
713 	 */
714 #ifdef SMP
715 	mtx_lock_spin(&smp_ipi_mtx);
716 #else
717 	spinlock_enter();
718 #endif
719 
720 	cpu_reset();
721 	/* NOTREACHED */ /* assuming reset worked */
722 }
723 
724 #if defined(WITNESS) || defined(INVARIANT_SUPPORT)
725 static int kassert_warn_only = 0;
726 #ifdef KDB
727 static int kassert_do_kdb = 0;
728 #endif
729 #ifdef KTR
730 static int kassert_do_ktr = 0;
731 #endif
732 static int kassert_do_log = 1;
733 static int kassert_log_pps_limit = 4;
734 static int kassert_log_mute_at = 0;
735 static int kassert_log_panic_at = 0;
736 static int kassert_suppress_in_panic = 0;
737 static int kassert_warnings = 0;
738 
739 SYSCTL_NODE(_debug, OID_AUTO, kassert, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
740     "kassert options");
741 
742 #ifdef KASSERT_PANIC_OPTIONAL
743 #define KASSERT_RWTUN	CTLFLAG_RWTUN
744 #else
745 #define KASSERT_RWTUN	CTLFLAG_RDTUN
746 #endif
747 
748 SYSCTL_INT(_debug_kassert, OID_AUTO, warn_only, KASSERT_RWTUN,
749     &kassert_warn_only, 0,
750     "KASSERT triggers a panic (0) or just a warning (1)");
751 
752 #ifdef KDB
753 SYSCTL_INT(_debug_kassert, OID_AUTO, do_kdb, KASSERT_RWTUN,
754     &kassert_do_kdb, 0, "KASSERT will enter the debugger");
755 #endif
756 
757 #ifdef KTR
758 SYSCTL_UINT(_debug_kassert, OID_AUTO, do_ktr, KASSERT_RWTUN,
759     &kassert_do_ktr, 0,
760     "KASSERT does a KTR, set this to the KTRMASK you want");
761 #endif
762 
763 SYSCTL_INT(_debug_kassert, OID_AUTO, do_log, KASSERT_RWTUN,
764     &kassert_do_log, 0,
765     "If warn_only is enabled, log (1) or do not log (0) assertion violations");
766 
767 SYSCTL_INT(_debug_kassert, OID_AUTO, warnings, CTLFLAG_RD | CTLFLAG_STATS,
768     &kassert_warnings, 0, "number of KASSERTs that have been triggered");
769 
770 SYSCTL_INT(_debug_kassert, OID_AUTO, log_panic_at, KASSERT_RWTUN,
771     &kassert_log_panic_at, 0, "max number of KASSERTS before we will panic");
772 
773 SYSCTL_INT(_debug_kassert, OID_AUTO, log_pps_limit, KASSERT_RWTUN,
774     &kassert_log_pps_limit, 0, "limit number of log messages per second");
775 
776 SYSCTL_INT(_debug_kassert, OID_AUTO, log_mute_at, KASSERT_RWTUN,
777     &kassert_log_mute_at, 0, "max number of KASSERTS to log");
778 
779 SYSCTL_INT(_debug_kassert, OID_AUTO, suppress_in_panic, KASSERT_RWTUN,
780     &kassert_suppress_in_panic, 0,
781     "KASSERTs will be suppressed while handling a panic");
782 #undef KASSERT_RWTUN
783 
784 static int kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS);
785 
786 SYSCTL_PROC(_debug_kassert, OID_AUTO, kassert,
787     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE | CTLFLAG_MPSAFE, NULL, 0,
788     kassert_sysctl_kassert, "I",
789     "set to trigger a test kassert");
790 
791 static int
kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)792 kassert_sysctl_kassert(SYSCTL_HANDLER_ARGS)
793 {
794 	int error, i;
795 
796 	error = sysctl_wire_old_buffer(req, sizeof(int));
797 	if (error == 0) {
798 		i = 0;
799 		error = sysctl_handle_int(oidp, &i, 0, req);
800 	}
801 	if (error != 0 || req->newptr == NULL)
802 		return (error);
803 	KASSERT(0, ("kassert_sysctl_kassert triggered kassert %d", i));
804 	return (0);
805 }
806 
807 #ifdef KASSERT_PANIC_OPTIONAL
808 /*
809  * Called by KASSERT, this decides if we will panic
810  * or if we will log via printf and/or ktr.
811  */
812 void
kassert_panic(const char * fmt,...)813 kassert_panic(const char *fmt, ...)
814 {
815 	static char buf[256];
816 	va_list ap;
817 
818 	va_start(ap, fmt);
819 	(void)vsnprintf(buf, sizeof(buf), fmt, ap);
820 	va_end(ap);
821 
822 	/*
823 	 * If we are suppressing secondary panics, log the warning but do not
824 	 * re-enter panic/kdb.
825 	 */
826 	if (KERNEL_PANICKED() && kassert_suppress_in_panic) {
827 		if (kassert_do_log) {
828 			printf("KASSERT failed: %s\n", buf);
829 #ifdef KDB
830 			if (trace_all_panics && trace_on_panic)
831 				kdb_backtrace();
832 #endif
833 		}
834 		return;
835 	}
836 
837 	/*
838 	 * panic if we're not just warning, or if we've exceeded
839 	 * kassert_log_panic_at warnings.
840 	 */
841 	if (!kassert_warn_only ||
842 	    (kassert_log_panic_at > 0 &&
843 	     kassert_warnings >= kassert_log_panic_at)) {
844 		va_start(ap, fmt);
845 		vpanic(fmt, ap);
846 		/* NORETURN */
847 	}
848 #ifdef KTR
849 	if (kassert_do_ktr)
850 		CTR0(ktr_mask, buf);
851 #endif /* KTR */
852 	/*
853 	 * log if we've not yet met the mute limit.
854 	 */
855 	if (kassert_do_log &&
856 	    (kassert_log_mute_at == 0 ||
857 	     kassert_warnings < kassert_log_mute_at)) {
858 		static  struct timeval lasterr;
859 		static  int curerr;
860 
861 		if (ppsratecheck(&lasterr, &curerr, kassert_log_pps_limit)) {
862 			printf("KASSERT failed: %s\n", buf);
863 			kdb_backtrace();
864 		}
865 	}
866 #ifdef KDB
867 	if (kassert_do_kdb) {
868 		kdb_enter(KDB_WHY_KASSERT, buf);
869 	}
870 #endif
871 	atomic_add_int(&kassert_warnings, 1);
872 }
873 #endif /* KASSERT_PANIC_OPTIONAL */
874 #endif
875 
876 /*
877  * Panic is called on unresolvable fatal errors.  It prints "panic: mesg",
878  * and then reboots.  If we are called twice, then we avoid trying to sync
879  * the disks as this often leads to recursive panics.
880  */
881 void
panic(const char * fmt,...)882 panic(const char *fmt, ...)
883 {
884 	va_list ap;
885 
886 	va_start(ap, fmt);
887 	vpanic(fmt, ap);
888 }
889 
890 void
vpanic(const char * fmt,va_list ap)891 vpanic(const char *fmt, va_list ap)
892 {
893 #ifdef SMP
894 	cpuset_t other_cpus;
895 #endif
896 	struct thread *td = curthread;
897 	int bootopt, newpanic;
898 	static char buf[256];
899 
900 	/*
901 	 * 'fmt' must not be NULL as it is put into 'panicstr' which is then
902 	 * used as a flag to detect if the kernel has panicked.  Also, although
903 	 * vsnprintf() supports a NULL 'fmt' argument, use a more informative
904 	 * message.
905 	 */
906 	if (fmt == NULL)
907 		fmt = "<no panic string!>";
908 
909 	spinlock_enter();
910 
911 #ifdef SMP
912 	/*
913 	 * stop_cpus_hard(other_cpus) should prevent multiple CPUs from
914 	 * concurrently entering panic.  Only the winner will proceed
915 	 * further.
916 	 */
917 	if (!KERNEL_PANICKED() && !kdb_active) {
918 		other_cpus = all_cpus;
919 		CPU_CLR(PCPU_GET(cpuid), &other_cpus);
920 		stop_cpus_hard(other_cpus);
921 	}
922 #endif
923 
924 	/*
925 	 * Ensure that the scheduler is stopped while panicking, even if panic
926 	 * has been entered from kdb.
927 	 */
928 	scheduler_stopped = true;
929 
930 	bootopt = RB_AUTOBOOT;
931 	newpanic = 0;
932 	if (KERNEL_PANICKED())
933 		bootopt |= RB_NOSYNC;
934 	else {
935 		bootopt |= RB_DUMP;
936 		panicstr = fmt;
937 		newpanic = 1;
938 	}
939 
940 	/* Unmute when panic */
941 	cn_mute = 0;
942 
943 	if (newpanic) {
944 		(void)vsnprintf(buf, sizeof(buf), fmt, ap);
945 		panicstr = buf;
946 		cngrab();
947 		printf("panic: %s\n", buf);
948 	} else {
949 		printf("panic: ");
950 		vprintf(fmt, ap);
951 		printf("\n");
952 	}
953 #ifdef SMP
954 	printf("cpuid = %d\n", PCPU_GET(cpuid));
955 #endif
956 	printf("time = %jd\n", (intmax_t )time_second);
957 #ifdef KDB
958 	if ((newpanic || trace_all_panics) && trace_on_panic)
959 		kdb_backtrace();
960 	if (debugger_on_panic)
961 		kdb_enter(KDB_WHY_PANIC, "panic");
962 	else if (!newpanic && debugger_on_recursive_panic)
963 		kdb_enter(KDB_WHY_PANIC, "re-panic");
964 #endif
965 	/*thread_lock(td); */
966 	td->td_flags |= TDF_INPANIC;
967 	/* thread_unlock(td); */
968 	if (!sync_on_panic)
969 		bootopt |= RB_NOSYNC;
970 	if (poweroff_on_panic)
971 		bootopt |= RB_POWEROFF;
972 	if (powercycle_on_panic)
973 		bootopt |= RB_POWERCYCLE;
974 	kern_reboot(bootopt);
975 }
976 
977 /*
978  * Support for poweroff delay.
979  *
980  * Please note that setting this delay too short might power off your machine
981  * before the write cache on your hard disk has been flushed, leading to
982  * soft-updates inconsistencies.
983  */
984 #ifndef POWEROFF_DELAY
985 # define POWEROFF_DELAY 5000
986 #endif
987 static int poweroff_delay = POWEROFF_DELAY;
988 
989 SYSCTL_INT(_kern_shutdown, OID_AUTO, poweroff_delay, CTLFLAG_RW,
990     &poweroff_delay, 0, "Delay before poweroff to write disk caches (msec)");
991 
992 static void
poweroff_wait(void * junk,int howto)993 poweroff_wait(void *junk, int howto)
994 {
995 
996 	if ((howto & (RB_POWEROFF | RB_POWERCYCLE)) == 0 || poweroff_delay <= 0)
997 		return;
998 	DELAY(poweroff_delay * 1000);
999 }
1000 
1001 /*
1002  * Some system processes (e.g. syncer) need to be stopped at appropriate
1003  * points in their main loops prior to a system shutdown, so that they
1004  * won't interfere with the shutdown process (e.g. by holding a disk buf
1005  * to cause sync to fail).  For each of these system processes, register
1006  * shutdown_kproc() as a handler for one of shutdown events.
1007  */
1008 static int kproc_shutdown_wait = 60;
1009 SYSCTL_INT(_kern_shutdown, OID_AUTO, kproc_shutdown_wait, CTLFLAG_RW,
1010     &kproc_shutdown_wait, 0, "Max wait time (sec) to stop for each process");
1011 
1012 void
kproc_shutdown(void * arg,int howto)1013 kproc_shutdown(void *arg, int howto)
1014 {
1015 	struct proc *p;
1016 	int error;
1017 
1018 	if (SCHEDULER_STOPPED())
1019 		return;
1020 
1021 	p = (struct proc *)arg;
1022 	printf("Waiting (max %d seconds) for system process `%s' to stop... ",
1023 	    kproc_shutdown_wait, p->p_comm);
1024 	error = kproc_suspend(p, kproc_shutdown_wait * hz);
1025 
1026 	if (error == EWOULDBLOCK)
1027 		printf("timed out\n");
1028 	else
1029 		printf("done\n");
1030 }
1031 
1032 void
kthread_shutdown(void * arg,int howto)1033 kthread_shutdown(void *arg, int howto)
1034 {
1035 	struct thread *td;
1036 	int error;
1037 
1038 	if (SCHEDULER_STOPPED())
1039 		return;
1040 
1041 	td = (struct thread *)arg;
1042 	printf("Waiting (max %d seconds) for system thread `%s' to stop... ",
1043 	    kproc_shutdown_wait, td->td_name);
1044 	error = kthread_suspend(td, kproc_shutdown_wait * hz);
1045 
1046 	if (error == EWOULDBLOCK)
1047 		printf("timed out\n");
1048 	else
1049 		printf("done\n");
1050 }
1051 
1052 static int
dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS)1053 dumpdevname_sysctl_handler(SYSCTL_HANDLER_ARGS)
1054 {
1055 	char buf[256];
1056 	struct dumperinfo *di;
1057 	struct sbuf sb;
1058 	int error;
1059 
1060 	error = sysctl_wire_old_buffer(req, 0);
1061 	if (error != 0)
1062 		return (error);
1063 
1064 	sbuf_new_for_sysctl(&sb, buf, sizeof(buf), req);
1065 
1066 	mtx_lock(&dumpconf_list_lk);
1067 	TAILQ_FOREACH(di, &dumper_configs, di_next) {
1068 		if (di != TAILQ_FIRST(&dumper_configs))
1069 			sbuf_putc(&sb, ',');
1070 		sbuf_cat(&sb, di->di_devname);
1071 	}
1072 	mtx_unlock(&dumpconf_list_lk);
1073 
1074 	error = sbuf_finish(&sb);
1075 	sbuf_delete(&sb);
1076 	return (error);
1077 }
1078 SYSCTL_PROC(_kern_shutdown, OID_AUTO, dumpdevname,
1079     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, &dumper_configs, 0,
1080     dumpdevname_sysctl_handler, "A",
1081     "Device(s) for kernel dumps");
1082 
1083 static int _dump_append(struct dumperinfo *di, void *virtual, size_t length);
1084 
1085 #ifdef EKCD
1086 static struct kerneldumpcrypto *
kerneldumpcrypto_create(size_t blocksize,uint8_t encryption,const uint8_t * key,uint32_t encryptedkeysize,const uint8_t * encryptedkey)1087 kerneldumpcrypto_create(size_t blocksize, uint8_t encryption,
1088     const uint8_t *key, uint32_t encryptedkeysize, const uint8_t *encryptedkey)
1089 {
1090 	struct kerneldumpcrypto *kdc;
1091 	struct kerneldumpkey *kdk;
1092 	uint32_t dumpkeysize;
1093 
1094 	dumpkeysize = roundup2(sizeof(*kdk) + encryptedkeysize, blocksize);
1095 	kdc = malloc(sizeof(*kdc) + dumpkeysize, M_EKCD, M_WAITOK | M_ZERO);
1096 
1097 	arc4rand(kdc->kdc_iv, sizeof(kdc->kdc_iv), 0);
1098 
1099 	kdc->kdc_encryption = encryption;
1100 	switch (kdc->kdc_encryption) {
1101 	case KERNELDUMP_ENC_AES_256_CBC:
1102 		if (rijndael_makeKey(&kdc->kdc_ki, DIR_ENCRYPT, 256, key) <= 0)
1103 			goto failed;
1104 		break;
1105 	case KERNELDUMP_ENC_CHACHA20:
1106 		chacha_keysetup(&kdc->kdc_chacha, key, 256);
1107 		break;
1108 	default:
1109 		goto failed;
1110 	}
1111 
1112 	kdc->kdc_dumpkeysize = dumpkeysize;
1113 	kdk = kdc->kdc_dumpkey;
1114 	kdk->kdk_encryption = kdc->kdc_encryption;
1115 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1116 	kdk->kdk_encryptedkeysize = htod32(encryptedkeysize);
1117 	memcpy(kdk->kdk_encryptedkey, encryptedkey, encryptedkeysize);
1118 
1119 	return (kdc);
1120 failed:
1121 	zfree(kdc, M_EKCD);
1122 	return (NULL);
1123 }
1124 
1125 static int
kerneldumpcrypto_init(struct kerneldumpcrypto * kdc)1126 kerneldumpcrypto_init(struct kerneldumpcrypto *kdc)
1127 {
1128 	uint8_t hash[SHA256_DIGEST_LENGTH];
1129 	SHA256_CTX ctx;
1130 	struct kerneldumpkey *kdk;
1131 	int error;
1132 
1133 	error = 0;
1134 
1135 	if (kdc == NULL)
1136 		return (0);
1137 
1138 	/*
1139 	 * When a user enters ddb it can write a crash dump multiple times.
1140 	 * Each time it should be encrypted using a different IV.
1141 	 */
1142 	SHA256_Init(&ctx);
1143 	SHA256_Update(&ctx, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1144 	SHA256_Final(hash, &ctx);
1145 	bcopy(hash, kdc->kdc_iv, sizeof(kdc->kdc_iv));
1146 
1147 	switch (kdc->kdc_encryption) {
1148 	case KERNELDUMP_ENC_AES_256_CBC:
1149 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1150 		    kdc->kdc_iv) <= 0) {
1151 			error = EINVAL;
1152 			goto out;
1153 		}
1154 		break;
1155 	case KERNELDUMP_ENC_CHACHA20:
1156 		chacha_ivsetup(&kdc->kdc_chacha, kdc->kdc_iv, NULL);
1157 		break;
1158 	default:
1159 		error = EINVAL;
1160 		goto out;
1161 	}
1162 
1163 	kdk = kdc->kdc_dumpkey;
1164 	memcpy(kdk->kdk_iv, kdc->kdc_iv, sizeof(kdk->kdk_iv));
1165 out:
1166 	explicit_bzero(hash, sizeof(hash));
1167 	return (error);
1168 }
1169 
1170 static uint32_t
kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto * kdc)1171 kerneldumpcrypto_dumpkeysize(const struct kerneldumpcrypto *kdc)
1172 {
1173 
1174 	if (kdc == NULL)
1175 		return (0);
1176 	return (kdc->kdc_dumpkeysize);
1177 }
1178 #endif /* EKCD */
1179 
1180 static struct kerneldumpcomp *
kerneldumpcomp_create(struct dumperinfo * di,uint8_t compression)1181 kerneldumpcomp_create(struct dumperinfo *di, uint8_t compression)
1182 {
1183 	struct kerneldumpcomp *kdcomp;
1184 	int format;
1185 
1186 	switch (compression) {
1187 	case KERNELDUMP_COMP_GZIP:
1188 		format = COMPRESS_GZIP;
1189 		break;
1190 	case KERNELDUMP_COMP_ZSTD:
1191 		format = COMPRESS_ZSTD;
1192 		break;
1193 	default:
1194 		return (NULL);
1195 	}
1196 
1197 	kdcomp = malloc(sizeof(*kdcomp), M_DUMPER, M_WAITOK | M_ZERO);
1198 	kdcomp->kdc_format = compression;
1199 	kdcomp->kdc_stream = compressor_init(kerneldumpcomp_write_cb,
1200 	    format, di->maxiosize, kerneldump_gzlevel, di);
1201 	if (kdcomp->kdc_stream == NULL) {
1202 		free(kdcomp, M_DUMPER);
1203 		return (NULL);
1204 	}
1205 	kdcomp->kdc_buf = malloc(di->maxiosize, M_DUMPER, M_WAITOK | M_NODUMP);
1206 	return (kdcomp);
1207 }
1208 
1209 static void
kerneldumpcomp_destroy(struct dumperinfo * di)1210 kerneldumpcomp_destroy(struct dumperinfo *di)
1211 {
1212 	struct kerneldumpcomp *kdcomp;
1213 
1214 	kdcomp = di->kdcomp;
1215 	if (kdcomp == NULL)
1216 		return;
1217 	compressor_fini(kdcomp->kdc_stream);
1218 	zfree(kdcomp->kdc_buf, M_DUMPER);
1219 	free(kdcomp, M_DUMPER);
1220 }
1221 
1222 /*
1223  * Free a dumper. Must not be present on global list.
1224  */
1225 void
dumper_destroy(struct dumperinfo * di)1226 dumper_destroy(struct dumperinfo *di)
1227 {
1228 
1229 	if (di == NULL)
1230 		return;
1231 
1232 	zfree(di->blockbuf, M_DUMPER);
1233 	kerneldumpcomp_destroy(di);
1234 #ifdef EKCD
1235 	zfree(di->kdcrypto, M_EKCD);
1236 #endif
1237 	zfree(di, M_DUMPER);
1238 }
1239 
1240 /*
1241  * Allocate and set up a new dumper from the provided template.
1242  */
1243 int
dumper_create(const struct dumperinfo * di_template,const char * devname,const struct diocskerneldump_arg * kda,struct dumperinfo ** dip)1244 dumper_create(const struct dumperinfo *di_template, const char *devname,
1245     const struct diocskerneldump_arg *kda, struct dumperinfo **dip)
1246 {
1247 	struct dumperinfo *newdi;
1248 	int error = 0;
1249 
1250 	if (dip == NULL)
1251 		return (EINVAL);
1252 
1253 	/* Allocate a new dumper */
1254 	newdi = malloc(sizeof(*newdi) + strlen(devname) + 1, M_DUMPER,
1255 	    M_WAITOK | M_ZERO);
1256 	memcpy(newdi, di_template, sizeof(*newdi));
1257 	newdi->blockbuf = NULL;
1258 	newdi->kdcrypto = NULL;
1259 	newdi->kdcomp = NULL;
1260 	strcpy(newdi->di_devname, devname);
1261 
1262 	if (kda->kda_encryption != KERNELDUMP_ENC_NONE) {
1263 #ifdef EKCD
1264 		newdi->kdcrypto = kerneldumpcrypto_create(newdi->blocksize,
1265 		    kda->kda_encryption, kda->kda_key,
1266 		    kda->kda_encryptedkeysize, kda->kda_encryptedkey);
1267 		if (newdi->kdcrypto == NULL) {
1268 			error = EINVAL;
1269 			goto cleanup;
1270 		}
1271 #else
1272 		error = EOPNOTSUPP;
1273 		goto cleanup;
1274 #endif
1275 	}
1276 	if (kda->kda_compression != KERNELDUMP_COMP_NONE) {
1277 #ifdef EKCD
1278 		/*
1279 		 * We can't support simultaneous unpadded block cipher
1280 		 * encryption and compression because there is no guarantee the
1281 		 * length of the compressed result is exactly a multiple of the
1282 		 * cipher block size.
1283 		 */
1284 		if (kda->kda_encryption == KERNELDUMP_ENC_AES_256_CBC) {
1285 			error = EOPNOTSUPP;
1286 			goto cleanup;
1287 		}
1288 #endif
1289 		newdi->kdcomp = kerneldumpcomp_create(newdi,
1290 		    kda->kda_compression);
1291 		if (newdi->kdcomp == NULL) {
1292 			error = EINVAL;
1293 			goto cleanup;
1294 		}
1295 	}
1296 	newdi->blockbuf = malloc(newdi->blocksize, M_DUMPER, M_WAITOK | M_ZERO);
1297 
1298 	*dip = newdi;
1299 	return (0);
1300 cleanup:
1301 	dumper_destroy(newdi);
1302 	return (error);
1303 }
1304 
1305 /*
1306  * Create a new dumper and register it in the global list.
1307  */
1308 int
dumper_insert(const struct dumperinfo * di_template,const char * devname,const struct diocskerneldump_arg * kda)1309 dumper_insert(const struct dumperinfo *di_template, const char *devname,
1310     const struct diocskerneldump_arg *kda)
1311 {
1312 	struct dumperinfo *newdi, *listdi;
1313 	bool inserted;
1314 	uint8_t index;
1315 	int error;
1316 
1317 	index = kda->kda_index;
1318 	MPASS(index != KDA_REMOVE && index != KDA_REMOVE_DEV &&
1319 	    index != KDA_REMOVE_ALL);
1320 
1321 	error = priv_check(curthread, PRIV_SETDUMPER);
1322 	if (error != 0)
1323 		return (error);
1324 
1325 	error = dumper_create(di_template, devname, kda, &newdi);
1326 	if (error != 0)
1327 		return (error);
1328 
1329 	/* Add the new configuration to the queue */
1330 	mtx_lock(&dumpconf_list_lk);
1331 	inserted = false;
1332 	TAILQ_FOREACH(listdi, &dumper_configs, di_next) {
1333 		if (index == 0) {
1334 			TAILQ_INSERT_BEFORE(listdi, newdi, di_next);
1335 			inserted = true;
1336 			break;
1337 		}
1338 		index--;
1339 	}
1340 	if (!inserted)
1341 		TAILQ_INSERT_TAIL(&dumper_configs, newdi, di_next);
1342 	mtx_unlock(&dumpconf_list_lk);
1343 
1344 	return (0);
1345 }
1346 
1347 #ifdef DDB
1348 void
dumper_ddb_insert(struct dumperinfo * newdi)1349 dumper_ddb_insert(struct dumperinfo *newdi)
1350 {
1351 	TAILQ_INSERT_HEAD(&dumper_configs, newdi, di_next);
1352 }
1353 
1354 void
dumper_ddb_remove(struct dumperinfo * di)1355 dumper_ddb_remove(struct dumperinfo *di)
1356 {
1357 	TAILQ_REMOVE(&dumper_configs, di, di_next);
1358 }
1359 #endif
1360 
1361 static bool
dumper_config_match(const struct dumperinfo * di,const char * devname,const struct diocskerneldump_arg * kda)1362 dumper_config_match(const struct dumperinfo *di, const char *devname,
1363     const struct diocskerneldump_arg *kda)
1364 {
1365 	if (kda->kda_index == KDA_REMOVE_ALL)
1366 		return (true);
1367 
1368 	if (strcmp(di->di_devname, devname) != 0)
1369 		return (false);
1370 
1371 	/*
1372 	 * Allow wildcard removal of configs matching a device on g_dev_orphan.
1373 	 */
1374 	if (kda->kda_index == KDA_REMOVE_DEV)
1375 		return (true);
1376 
1377 	if (di->kdcomp != NULL) {
1378 		if (di->kdcomp->kdc_format != kda->kda_compression)
1379 			return (false);
1380 	} else if (kda->kda_compression != KERNELDUMP_COMP_NONE)
1381 		return (false);
1382 #ifdef EKCD
1383 	if (di->kdcrypto != NULL) {
1384 		if (di->kdcrypto->kdc_encryption != kda->kda_encryption)
1385 			return (false);
1386 		/*
1387 		 * Do we care to verify keys match to delete?  It seems weird
1388 		 * to expect multiple fallback dump configurations on the same
1389 		 * device that only differ in crypto key.
1390 		 */
1391 	} else
1392 #endif
1393 		if (kda->kda_encryption != KERNELDUMP_ENC_NONE)
1394 			return (false);
1395 
1396 	return (true);
1397 }
1398 
1399 /*
1400  * Remove and free the requested dumper(s) from the global list.
1401  */
1402 int
dumper_remove(const char * devname,const struct diocskerneldump_arg * kda)1403 dumper_remove(const char *devname, const struct diocskerneldump_arg *kda)
1404 {
1405 	struct dumperinfo *di, *sdi;
1406 	bool found;
1407 	int error;
1408 
1409 	error = priv_check(curthread, PRIV_SETDUMPER);
1410 	if (error != 0)
1411 		return (error);
1412 
1413 	/*
1414 	 * Try to find a matching configuration, and kill it.
1415 	 *
1416 	 * NULL 'kda' indicates remove any configuration matching 'devname',
1417 	 * which may remove multiple configurations in atypical configurations.
1418 	 */
1419 	found = false;
1420 	mtx_lock(&dumpconf_list_lk);
1421 	TAILQ_FOREACH_SAFE(di, &dumper_configs, di_next, sdi) {
1422 		if (dumper_config_match(di, devname, kda)) {
1423 			found = true;
1424 			TAILQ_REMOVE(&dumper_configs, di, di_next);
1425 			dumper_destroy(di);
1426 		}
1427 	}
1428 	mtx_unlock(&dumpconf_list_lk);
1429 
1430 	/* Only produce ENOENT if a more targeted match didn't match. */
1431 	if (!found && kda->kda_index == KDA_REMOVE)
1432 		return (ENOENT);
1433 	return (0);
1434 }
1435 
1436 static int
dump_check_bounds(struct dumperinfo * di,off_t offset,size_t length)1437 dump_check_bounds(struct dumperinfo *di, off_t offset, size_t length)
1438 {
1439 
1440 	if (di->mediasize > 0 && length != 0 && (offset < di->mediaoffset ||
1441 	    offset - di->mediaoffset + length > di->mediasize)) {
1442 		if (di->kdcomp != NULL && offset >= di->mediaoffset) {
1443 			printf(
1444 		    "Compressed dump failed to fit in device boundaries.\n");
1445 			return (E2BIG);
1446 		}
1447 
1448 		printf("Attempt to write outside dump device boundaries.\n"
1449 	    "offset(%jd), mediaoffset(%jd), length(%ju), mediasize(%jd).\n",
1450 		    (intmax_t)offset, (intmax_t)di->mediaoffset,
1451 		    (uintmax_t)length, (intmax_t)di->mediasize);
1452 		return (ENOSPC);
1453 	}
1454 	if (length % di->blocksize != 0) {
1455 		printf("Attempt to write partial block of length %ju.\n",
1456 		    (uintmax_t)length);
1457 		return (EINVAL);
1458 	}
1459 	if (offset % di->blocksize != 0) {
1460 		printf("Attempt to write at unaligned offset %jd.\n",
1461 		    (intmax_t)offset);
1462 		return (EINVAL);
1463 	}
1464 
1465 	return (0);
1466 }
1467 
1468 #ifdef EKCD
1469 static int
dump_encrypt(struct kerneldumpcrypto * kdc,uint8_t * buf,size_t size)1470 dump_encrypt(struct kerneldumpcrypto *kdc, uint8_t *buf, size_t size)
1471 {
1472 
1473 	switch (kdc->kdc_encryption) {
1474 	case KERNELDUMP_ENC_AES_256_CBC:
1475 		if (rijndael_blockEncrypt(&kdc->kdc_ci, &kdc->kdc_ki, buf,
1476 		    8 * size, buf) <= 0) {
1477 			return (EIO);
1478 		}
1479 		if (rijndael_cipherInit(&kdc->kdc_ci, MODE_CBC,
1480 		    buf + size - 16 /* IV size for AES-256-CBC */) <= 0) {
1481 			return (EIO);
1482 		}
1483 		break;
1484 	case KERNELDUMP_ENC_CHACHA20:
1485 		chacha_encrypt_bytes(&kdc->kdc_chacha, buf, buf, size);
1486 		break;
1487 	default:
1488 		return (EINVAL);
1489 	}
1490 
1491 	return (0);
1492 }
1493 
1494 /* Encrypt data and call dumper. */
1495 static int
dump_encrypted_write(struct dumperinfo * di,void * virtual,off_t offset,size_t length)1496 dump_encrypted_write(struct dumperinfo *di, void *virtual, off_t offset,
1497     size_t length)
1498 {
1499 	static uint8_t buf[KERNELDUMP_BUFFER_SIZE];
1500 	struct kerneldumpcrypto *kdc;
1501 	int error;
1502 	size_t nbytes;
1503 
1504 	kdc = di->kdcrypto;
1505 
1506 	while (length > 0) {
1507 		nbytes = MIN(length, sizeof(buf));
1508 		bcopy(virtual, buf, nbytes);
1509 
1510 		if (dump_encrypt(kdc, buf, nbytes) != 0)
1511 			return (EIO);
1512 
1513 		error = dump_write(di, buf, offset, nbytes);
1514 		if (error != 0)
1515 			return (error);
1516 
1517 		offset += nbytes;
1518 		virtual = (void *)((uint8_t *)virtual + nbytes);
1519 		length -= nbytes;
1520 	}
1521 
1522 	return (0);
1523 }
1524 #endif /* EKCD */
1525 
1526 static int
kerneldumpcomp_write_cb(void * base,size_t length,off_t offset,void * arg)1527 kerneldumpcomp_write_cb(void *base, size_t length, off_t offset, void *arg)
1528 {
1529 	struct dumperinfo *di;
1530 	size_t resid, rlength;
1531 	int error;
1532 
1533 	di = arg;
1534 
1535 	if (length % di->blocksize != 0) {
1536 		/*
1537 		 * This must be the final write after flushing the compression
1538 		 * stream. Write as many full blocks as possible and stash the
1539 		 * residual data in the dumper's block buffer. It will be
1540 		 * padded and written in dump_finish().
1541 		 */
1542 		rlength = rounddown(length, di->blocksize);
1543 		if (rlength != 0) {
1544 			error = _dump_append(di, base, rlength);
1545 			if (error != 0)
1546 				return (error);
1547 		}
1548 		resid = length - rlength;
1549 		memmove(di->blockbuf, (uint8_t *)base + rlength, resid);
1550 		bzero((uint8_t *)di->blockbuf + resid, di->blocksize - resid);
1551 		di->kdcomp->kdc_resid = resid;
1552 		return (EAGAIN);
1553 	}
1554 	return (_dump_append(di, base, length));
1555 }
1556 
1557 /*
1558  * Write kernel dump headers at the beginning and end of the dump extent.
1559  * Write the kernel dump encryption key after the leading header if we were
1560  * configured to do so.
1561  */
1562 static int
dump_write_headers(struct dumperinfo * di,struct kerneldumpheader * kdh)1563 dump_write_headers(struct dumperinfo *di, struct kerneldumpheader *kdh)
1564 {
1565 #ifdef EKCD
1566 	struct kerneldumpcrypto *kdc;
1567 #endif
1568 	void *buf;
1569 	size_t hdrsz;
1570 	uint64_t extent;
1571 	uint32_t keysize;
1572 	int error;
1573 
1574 	hdrsz = sizeof(*kdh);
1575 	if (hdrsz > di->blocksize)
1576 		return (ENOMEM);
1577 
1578 #ifdef EKCD
1579 	kdc = di->kdcrypto;
1580 	keysize = kerneldumpcrypto_dumpkeysize(kdc);
1581 #else
1582 	keysize = 0;
1583 #endif
1584 
1585 	/*
1586 	 * If the dump device has special handling for headers, let it take care
1587 	 * of writing them out.
1588 	 */
1589 	if (di->dumper_hdr != NULL)
1590 		return (di->dumper_hdr(di, kdh));
1591 
1592 	if (hdrsz == di->blocksize)
1593 		buf = kdh;
1594 	else {
1595 		buf = di->blockbuf;
1596 		memset(buf, 0, di->blocksize);
1597 		memcpy(buf, kdh, hdrsz);
1598 	}
1599 
1600 	extent = dtoh64(kdh->dumpextent);
1601 #ifdef EKCD
1602 	if (kdc != NULL) {
1603 		error = dump_write(di, kdc->kdc_dumpkey,
1604 		    di->mediaoffset + di->mediasize - di->blocksize - extent -
1605 		    keysize, keysize);
1606 		if (error != 0)
1607 			return (error);
1608 	}
1609 #endif
1610 
1611 	error = dump_write(di, buf,
1612 	    di->mediaoffset + di->mediasize - 2 * di->blocksize - extent -
1613 	    keysize, di->blocksize);
1614 	if (error == 0)
1615 		error = dump_write(di, buf, di->mediaoffset + di->mediasize -
1616 		    di->blocksize, di->blocksize);
1617 	return (error);
1618 }
1619 
1620 /*
1621  * Don't touch the first SIZEOF_METADATA bytes on the dump device.  This is to
1622  * protect us from metadata and metadata from us.
1623  */
1624 #define	SIZEOF_METADATA		(64 * 1024)
1625 
1626 /*
1627  * Do some preliminary setup for a kernel dump: initialize state for encryption,
1628  * if requested, and make sure that we have enough space on the dump device.
1629  *
1630  * We set things up so that the dump ends before the last sector of the dump
1631  * device, at which the trailing header is written.
1632  *
1633  *     +-----------+------+-----+----------------------------+------+
1634  *     |           | lhdr | key |    ... kernel dump ...     | thdr |
1635  *     +-----------+------+-----+----------------------------+------+
1636  *                   1 blk  opt <------- dump extent --------> 1 blk
1637  *
1638  * Dumps written using dump_append() start at the beginning of the extent.
1639  * Uncompressed dumps will use the entire extent, but compressed dumps typically
1640  * will not. The true length of the dump is recorded in the leading and trailing
1641  * headers once the dump has been completed.
1642  *
1643  * The dump device may provide a callback, in which case it will initialize
1644  * dumpoff and take care of laying out the headers.
1645  */
1646 int
dump_start(struct dumperinfo * di,struct kerneldumpheader * kdh)1647 dump_start(struct dumperinfo *di, struct kerneldumpheader *kdh)
1648 {
1649 #ifdef EKCD
1650 	struct kerneldumpcrypto *kdc;
1651 #endif
1652 	void *key;
1653 	uint64_t dumpextent, span;
1654 	uint32_t keysize;
1655 	int error;
1656 
1657 #ifdef EKCD
1658 	/* Send the key before the dump so a partial dump is still usable. */
1659 	kdc = di->kdcrypto;
1660 	error = kerneldumpcrypto_init(kdc);
1661 	if (error != 0)
1662 		return (error);
1663 	keysize = kerneldumpcrypto_dumpkeysize(kdc);
1664 	key = keysize > 0 ? kdc->kdc_dumpkey : NULL;
1665 #else
1666 	error = 0;
1667 	keysize = 0;
1668 	key = NULL;
1669 #endif
1670 
1671 	if (di->dumper_start != NULL) {
1672 		error = di->dumper_start(di, key, keysize);
1673 	} else {
1674 		dumpextent = dtoh64(kdh->dumpextent);
1675 		span = SIZEOF_METADATA + dumpextent + 2 * di->blocksize +
1676 		    keysize;
1677 		if (di->mediasize < span) {
1678 			if (di->kdcomp == NULL)
1679 				return (E2BIG);
1680 
1681 			/*
1682 			 * We don't yet know how much space the compressed dump
1683 			 * will occupy, so try to use the whole swap partition
1684 			 * (minus the first 64KB) in the hope that the
1685 			 * compressed dump will fit. If that doesn't turn out to
1686 			 * be enough, the bounds checking in dump_write()
1687 			 * will catch us and cause the dump to fail.
1688 			 */
1689 			dumpextent = di->mediasize - span + dumpextent;
1690 			kdh->dumpextent = htod64(dumpextent);
1691 		}
1692 
1693 		/*
1694 		 * The offset at which to begin writing the dump.
1695 		 */
1696 		di->dumpoff = di->mediaoffset + di->mediasize - di->blocksize -
1697 		    dumpextent;
1698 	}
1699 	di->origdumpoff = di->dumpoff;
1700 	return (error);
1701 }
1702 
1703 static int
_dump_append(struct dumperinfo * di,void * virtual,size_t length)1704 _dump_append(struct dumperinfo *di, void *virtual, size_t length)
1705 {
1706 	int error;
1707 
1708 #ifdef EKCD
1709 	if (di->kdcrypto != NULL)
1710 		error = dump_encrypted_write(di, virtual, di->dumpoff, length);
1711 	else
1712 #endif
1713 		error = dump_write(di, virtual, di->dumpoff, length);
1714 	if (error == 0)
1715 		di->dumpoff += length;
1716 	return (error);
1717 }
1718 
1719 /*
1720  * Write to the dump device starting at dumpoff. When compression is enabled,
1721  * writes to the device will be performed using a callback that gets invoked
1722  * when the compression stream's output buffer is full.
1723  */
1724 int
dump_append(struct dumperinfo * di,void * virtual,size_t length)1725 dump_append(struct dumperinfo *di, void *virtual, size_t length)
1726 {
1727 	void *buf;
1728 
1729 	if (di->kdcomp != NULL) {
1730 		/* Bounce through a buffer to avoid CRC errors. */
1731 		if (length > di->maxiosize)
1732 			return (EINVAL);
1733 		buf = di->kdcomp->kdc_buf;
1734 		memmove(buf, virtual, length);
1735 		return (compressor_write(di->kdcomp->kdc_stream, buf, length));
1736 	}
1737 	return (_dump_append(di, virtual, length));
1738 }
1739 
1740 /*
1741  * Write to the dump device at the specified offset.
1742  */
1743 int
dump_write(struct dumperinfo * di,void * virtual,off_t offset,size_t length)1744 dump_write(struct dumperinfo *di, void *virtual, off_t offset, size_t length)
1745 {
1746 	int error;
1747 
1748 	error = dump_check_bounds(di, offset, length);
1749 	if (error != 0)
1750 		return (error);
1751 	return (di->dumper(di->priv, virtual, offset, length));
1752 }
1753 
1754 /*
1755  * Perform kernel dump finalization: flush the compression stream, if necessary,
1756  * write the leading and trailing kernel dump headers now that we know the true
1757  * length of the dump, and optionally write the encryption key following the
1758  * leading header.
1759  */
1760 int
dump_finish(struct dumperinfo * di,struct kerneldumpheader * kdh)1761 dump_finish(struct dumperinfo *di, struct kerneldumpheader *kdh)
1762 {
1763 	int error;
1764 
1765 	if (di->kdcomp != NULL) {
1766 		error = compressor_flush(di->kdcomp->kdc_stream);
1767 		if (error == EAGAIN) {
1768 			/* We have residual data in di->blockbuf. */
1769 			error = _dump_append(di, di->blockbuf, di->blocksize);
1770 			if (error == 0)
1771 				/* Compensate for _dump_append()'s adjustment. */
1772 				di->dumpoff -= di->blocksize - di->kdcomp->kdc_resid;
1773 			di->kdcomp->kdc_resid = 0;
1774 		}
1775 		if (error != 0)
1776 			return (error);
1777 
1778 		/*
1779 		 * We now know the size of the compressed dump, so update the
1780 		 * header accordingly and recompute parity.
1781 		 */
1782 		kdh->dumplength = htod64(di->dumpoff - di->origdumpoff);
1783 		kdh->parity = 0;
1784 		kdh->parity = kerneldump_parity(kdh);
1785 
1786 		compressor_reset(di->kdcomp->kdc_stream);
1787 	}
1788 
1789 	error = dump_write_headers(di, kdh);
1790 	if (error != 0)
1791 		return (error);
1792 
1793 	(void)dump_write(di, NULL, 0, 0);
1794 	return (0);
1795 }
1796 
1797 void
dump_init_header(const struct dumperinfo * di,struct kerneldumpheader * kdh,const char * magic,uint32_t archver,uint64_t dumplen)1798 dump_init_header(const struct dumperinfo *di, struct kerneldumpheader *kdh,
1799     const char *magic, uint32_t archver, uint64_t dumplen)
1800 {
1801 	size_t dstsize;
1802 
1803 	bzero(kdh, sizeof(*kdh));
1804 	strlcpy(kdh->magic, magic, sizeof(kdh->magic));
1805 	strlcpy(kdh->architecture, MACHINE_ARCH, sizeof(kdh->architecture));
1806 	kdh->version = htod32(KERNELDUMPVERSION);
1807 	kdh->architectureversion = htod32(archver);
1808 	kdh->dumplength = htod64(dumplen);
1809 	kdh->dumpextent = kdh->dumplength;
1810 	kdh->dumptime = htod64(time_second);
1811 #ifdef EKCD
1812 	kdh->dumpkeysize = htod32(kerneldumpcrypto_dumpkeysize(di->kdcrypto));
1813 #else
1814 	kdh->dumpkeysize = 0;
1815 #endif
1816 	kdh->blocksize = htod32(di->blocksize);
1817 	strlcpy(kdh->hostname, prison0.pr_hostname, sizeof(kdh->hostname));
1818 	dstsize = sizeof(kdh->versionstring);
1819 	if (strlcpy(kdh->versionstring, version, dstsize) >= dstsize)
1820 		kdh->versionstring[dstsize - 2] = '\n';
1821 	if (panicstr != NULL)
1822 		strlcpy(kdh->panicstring, panicstr, sizeof(kdh->panicstring));
1823 	if (di->kdcomp != NULL)
1824 		kdh->compression = di->kdcomp->kdc_format;
1825 	kdh->parity = kerneldump_parity(kdh);
1826 }
1827 
1828 #ifdef DDB
DB_SHOW_COMMAND_FLAGS(panic,db_show_panic,DB_CMD_MEMSAFE)1829 DB_SHOW_COMMAND_FLAGS(panic, db_show_panic, DB_CMD_MEMSAFE)
1830 {
1831 
1832 	if (panicstr == NULL)
1833 		db_printf("panicstr not set\n");
1834 	else
1835 		db_printf("panic: %s\n", panicstr);
1836 }
1837 #endif
1838